A joint deep learning model enables simultaneous batch effect correction, denoising, and clustering in single-cell...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables high-resolution biological discoveries.
• Batch effects are a significant challenge in scRNA-seq analysis, particularly with human tissues.
• Existing methods often correct batch effects in low-dimensional spaces, leaving gene expression susceptible to noise.

Purpose of the Study:

• To develop a joint deep learning model for simultaneous clustering, denoising, and batch effect correction in scRNA-seq data.
• To address the limitations of existing methods in correcting batch effects in the gene expression space.
• To improve the accuracy and interpretability of downstream analyses, including marker gene identification and trajectory inference.

Main Methods:

• Developed CarDEC, a joint deep learning model.
• Simultaneously performs clustering, denoising, and batch effect correction.
• Corrects batch effects in both embedding and gene expression spaces.

Main Results:

• CarDEC outperforms existing methods like Scanorama, DCA + Combat, scVI, and MNN in comprehensive evaluations.
• Denoising with CarDEC enhances the signal from non-highly variable genes for clustering.
• Trajectory analysis using CarDEC-corrected data reveals obscured marker genes and transcription factors.

Conclusions:

• CarDEC offers robust batch effect correction in both embedding and gene expression spaces for scRNA-seq data.
• The model significantly improves the information content and analytical power of scRNA-seq datasets.
• CarDEC's computational efficiency makes it suitable for large-scale scRNA-seq studies.